期刊文献+

一种基于多尺度角点检测的语义分割网络

A Precise Image Semantice Segmentation Model Based on Multi-scale Corner Detection
在线阅读 下载PDF
收藏 分享 导出
摘要 为了实现更精确的语义分割,提出了一种目标全局解析网络(object global parsing network,OGPNet)。首先,基于卷积特征金字塔构造了一个多尺度角点检测器,检测不同尺度特征图上目标的关键点信息;其次,提出了一种多尺度联合池算法将获得的多尺度角点进行融合;最后,将组归一化(Grounp Normalization,GN)方法引入到该分割网络训练中以提升网络训练和收敛速度。OGPNet在Pascal VOC 2012数据集和Cityscapes数据集的分割结果的mIoU评价分别达到了78.5%和67.6%。且实验证明,相对于现有的一些语义分割网络,由OGPNet分割出的目标具有更完整的轮廓,且分割结果的视觉质量更好。 In order to achieve more precise semantic segmentation,an object global parsing network(OGPNet)is presented in this pa per.Firstly,a multi-scale corner detector is constructed Based on the convolution feature pyramid to identify the corner information of objects in feature maps which with different scales.Secondly,a multi-scale joint pooling algorithm is proposed to fuse the obtained multi-scale corner.Finally,Group Normalization(GN)method is introduced in training OGPNet to accelerate the convergence speed of this network.The Pascal VOC 2012 dataset and Cityscapes dataset are used to train and test OGPNet.And the mIoU of the segmentation results reached 78.5%and 67.6%.Experiments also indicate that compared with some existing semantic segmentation networks,the ob jects segmented by OGPNet have more complete contour,and the visual quality of the segmentation results is better.
作者 罗晖 芦春雨 郑翔文 LUO Hui;LU Chun-yu;ZHENG Xiang-wen(Department of Informantion Engineering East Chian Jiao Tong University,Nanchang 330013,China)
出处 《电脑知识与技术:学术版》 2019年第11X期206-210,共5页 Computer Knowledge and Technology
关键词 语义分割 多尺度 角点检测 联合池化 组归一化 Semantic segmentation Multi-scale Corner Detector Group normalization Joint pooling.
  • 相关文献
论文智能改写系统
维普数据出版直通车
投稿分析
职称考试

相关作者

内容加载中请稍等...

相关机构

内容加载中请稍等...

相关主题

内容加载中请稍等...

浏览历史

内容加载中请稍等...
;
使用帮助 返回顶部 意见反馈